VisDA 2013 solicits contributions on all topics related to visual dataset bias and visual domain adaptation, with special interests in:
(1) Fundamental studies on visual dataset bias and domain shifts. Physical and statistical characterizations of dataset bias and domain shifts; measuring distribution mismatch and generalization bounds for visual data; visual-prior-guided learning; integration of physical and statistical models.
(2) Novel paradigms accounting for various cross-domain constraints. Feature co-occurrences between domains; semi-supervised and unsupervised adaptations; category and instance level constraints; heterogeneous domains; multiple source domains; transferring unseen categories; negative transfer.
(3) Adapting vision-specific representations and models. Adapting detection, segmentation, reconstruction, and tracking algorithms to new domains; adapting representations of imaging processes, shape and deformations, pictorial structure and graphs, random fields, and visual dynamics to new domains.
(4) Efficient adapting large-scale visual data. Scalable algorithms for adaptation between large datasets, incremental adaptation.
(5) Development of rigorous multi-domain datasets, challenges, and evaluation paradigms.